基于改进傅里叶神经网络的多关节机器人实时负载辨识方法  

Real-time load identification method for multi-joint robots based on improved Fourier neural

作　　者：岳夏[1] 李志滨 张春良[1] 王亚东 王宇华 龙尚斌 YUE Xia;LI Zhibin;ZHANG Chunliang;WANG Yadong;WANG Yuhua;LONG Shangbin(School of Mechanical and Electrical Engineering,Guangzhou University,Guangzhou 510006,China)

机构地区：[1]广州大学机械与电气工程学院,广州510006

出　　处：《振动与冲击》2025年第5期314-322,共9页Journal of Vibration and Shock

基　　金：广东华中科技大学工业技术研究院/广东省制造装备数字化重点实验室开放基金(2023B1212060012);国家自然科学基金面上项目(52275097)。

摘　　要：关节式机器人应用于各类生产环节,对负载进行实时监测是确保机器人安全运行的前提。但在某些特殊场景下无法直接测量负载,通常使用动力学方法间接求解,由于其非线性特性明显且模型参数的不确定性,负载识别的精度与效率一直不高。因此该研究基于傅里叶神经网络提出了一种改进模型来实现负载辨识,以提高系统负载参数的预测精度与时效性。所提方法利用傅里叶神经网络中的卷积与频域截断机制快速获取特征信号,与前馈神经网络的输出结果进行数据融合得到辨识结果。所提方法相比动力学模型求解方法精度更高、计算速度更快,仅需学习预测范围内几个相间的样本集,就可识别预测范围内的任意结果,泛化能力好。同时进行网络敏感参数的分析,并与成熟神经网络算法进行性能比较。该方法将两种神经网络模型进行协同配合,能有效识别高维数据中的不同特征集,从而实现参数辨识,为复杂非线性系统的参数识别提供参考。Joint robots are applied in various production links,and real-time monitoring of loads is a prerequisite to ensure safe operation of robots.However,in some special scenarios,it is not possible to directly measure loads,and dynamic methods are usually used to indirectly solve them.Due to their obvious nonlinear characteristics and the uncertainty of model parameters,the accuracy and efficiency of load identification always are low.Here,an improved model based on Fourier neural network was proposed to realize load identification,and improve the prediction accuracy and timeliness of system load parameters.The proposed method could use convolution and frequency domain truncation mechanism in Fourier neural network to quickly obtain feature signals,which then were fused with output results of feedforward neural network to obtain identification results.It was shown that the proposed method had higher accuracy and faster calculation speed compared to the dynamic model solving method;it only need to learn a few interphase sample sets within the prediction range to identify any result within the prediction range,and had good generalization ability;it could simultaneously analyze network sensitive parameters and perform performance comparison with mature neural network algorithms.This method can collaborate two types of neural network models to effectively identify different feature sets in high-dimensional data,realize parametric identification and provide a reference for parametric identification of complex nonlinear systems.

关 键 词：工业机器人 傅里叶神经网络 动力学 实时 负载识别 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TP242[自动化与计算机技术—控制科学与工程]